Abstract: Object localization aims to estimate the position and orientation of the object in 3D space. In static scenes, it is equivalent to camera tracking, which has a wide range of applications. In this paper, we propose a contour-based approach to localize texture-less irregular objects with known 3D models. The target object is first extracted from the input image via image segmentation, the contour is then matched with rendered 3D model with given position and orientation parameters. The matching error can be expressed as a function of the position and orientation parameters. Since the function cannot be analytically expressed and solved, we calculate the matching score and derivative by discrete sampling, the optimal score and derivative parameters can be solved efficiently via LM（Levenberg-Marquardt） solver. Experimental results show that this method can converge quickly, and can achieve very high accuracy and robustness.